Population genetic structure and its implications for adaptive variation in memory and the hippocampus on a continental scale in food-caching black-capped chickadees V. V. PRAVOSUDOV,* T . C. ROTH II , *† M. L. FORISTER,* L . D. LADAGE,* T . M. BURG,‡ M. J . BRAUN§ and B. S . DAVIDSON§ *Department of Biology, University of Nevada Reno, MS314, Reno, NV, 89557, USA, †Department of Biology, Kenyon College, Gambier, OH, 43022, USA, ‡Department of Biological Sciences, University of Lethbridge, Lethbridge, AB, T1K 3M4, Canada, §Department of Vertebrate Zoology, National Museum of Natural History, Smithsonian Institution, Washington, DC, 20560, USA Abstract Food-caching birds rely on stored food to survive the winter, and spatial memory has been shown to be critical in successful cache recovery. Both spatial memory and the hippocampus, an area of the brain involved in spatial memory, exhibit significant geo- graphic variation linked to climate-based environmental harshness and the potential reliance on food caches for survival. Such geographic variation has been suggested to have a heritable basis associated with differential selection. Here, we ask whether pop- ulation genetic differentiation and potential isolation among multiple populations of food-caching black-capped chickadees is associated with differences in memory and hippocampal morphology by exploring population genetic structure within and among groups of populations that are divergent to different degrees in hippocampal morphol- ogy. Using mitochondrial DNA and 583 AFLP loci, we found that population diver- gence in hippocampal morphology is not significantly associated with neutral genetic divergence or geographic distance, but instead is significantly associated with differ- ences in winter climate. These results are consistent with variation in a history of natu- ral selection on memory and hippocampal morphology that creates and maintains differences in these traits regardless of population genetic structure and likely associ- ated gene flow. Keywords: adaptation, chickadee, food caching, gene flow, hippocampus, memory, population structure Received 22 March 2012; revision received 18 June 2012; accepted 20 June 2012 Introduction The classic view of natural selection is that, given varia- tion in heritable traits, strong directional selection should produce adaptive changes in these traits (Orr 2005). This appears to be the case in scatter-hoarding species, where directional selection has been suggested to produce specializations in memory and the brain. More specifically, the adaptive specialization hypothesis has been proposed to explain reported differences in the size of the hippocampus, an area of the brain involved in spatial memory, between food-caching and noncaching species (Krebs et al. 1989; Sherry et al. 1989; Pravosudov & Smulders 2010). Scatter-hoarding species cache large numbers of individual food items scattered over their home ranges and retrieve them during the winter when naturally available food is limited and unpredictable (Pravosudov & Smulders 2010). Individu- als of some species, especially corvids and parids, have been reported to cache tens of thousands of food items during a single year (Praavosudov & Smulders 2010; Vander Wall 1990), which suggests a critical role of these caches in individual fitness. Successful retrieval of Correspondence: Vladimir V. Pravosudov, Fax: 775 7841302; E-mail: vpravosu@unr.edu Published 2012. This article is a US Government work and is in the public domain in the USA. Molecular Ecology (2012) 21, 4486–4497 doi: 10.1111/j.1365-294X.2012.05721.x food caches should be especially important for fitness, and therefore, selection is likely to act on the mecha- nisms used for cache retrieval. Individuals with traits allowing for more successful cache retrieval should have a selective advantage, which should in turn result in the evolution of these traits. It has been shown that food-caching animals rely, at least in part, on spatial memory to recover previously cached food, and thus, better spatial memory should result in more successful cache recovery (Shettleworth 1995; Pravosudov & Smul- ders 2010). The hippocampus is involved in spatial memory function (Sherry & Vaccarino 1989; Hampton & Shettleworth 1996); hence, selection pressure on spa- tial memory needed to recover food caches may be expected to result in adaptive changes in hippocampal morphology and size in food-caching animals (Krebs et al. 1989; Sherry et al. 1989; Pravosudov & Smulders 2010). We have previously examined how reliance on food caches may affect the association between spatial mem- ory and the hippocampus by using the black-capped chickadee (Poecile atricapillus), a food-caching species with an extremely large range over the North American continent encompassing a great variety of environmen- tal conditions (Pravosudov & Clayton 2002; Roth & Pravosudov 2009; Chancellor et al. 2011; Roth et al. 2011, 2012). Our previous studies were based on the hypothesis that winter climate is the main factor affect- ing selection on spatial memory in these small passerine birds. Colder winter temperatures demand more energy to sustain higher metabolic requirements, while both low temperatures and abundant snow fall and cover likely make naturally available food limited and unpre- dictable, all of which should increase dependence on cached food for survival (Pravosudov & Clayton 2002; Roth & Pravosudov 2009; Roth et al. 2011). Our previous studies compared chickadees from 10 locations spread over a large continental scale (Alaska, Fairbanks (AKF), Alaska, Anchorage (AKA), British Columbia (BC), Washington State (WA), Montana (MO), Minnesota (MN), Colorado (CO), Kansas (KS), Iowa (IA) and Maine (ME), Fig. 1) and reported that chickadees from more harsh environments (AKA, AKF, ME, MN) had relatively larger hippocampi containing more hippocampal neurons compared to birds from more moderate climates (Pravosudov & Clayton 2002; Roth & Pravosudov 2009; Roth et al. 2011). These data on hippocampal morphology are summarized in Table 1. We also conducted a common garden experi- ment with birds from the two extreme populations (AKA and KS; Fig. 1) to test whether previously found differences may be due to plasticity associated with environmental influences. Our data strongly suggested the role of inheritance as the birds hand-reared in a controlled laboratory environment exhibited the same differences in spatial memory, total number of hippo- campal neurons and intensity of hippocampal neuro- genesis as the birds sampled from the wild (Roth et al. 2012). Even though we used only the two most extreme populations for the common garden study, it is plau- sible to assume that the same mechanism (most likely genetically based) may be involved in producing the reported differences across all of our sampled popula- tions and that this variation may be, at least in part, due to differential selection on spatial memory. Our data have shown that variation in the number of hip- pocampal neurons in black-capped chickadees has a heritable component, as birds reared and maintained in laboratory conditions and chickadees sampled directly in the wild had statistically indistinguishable numbers of hippocampal neurons despite enormous differences between natural and laboratory environ- ments (Roth et al. 2012). These data suggest that a significant component of the variation in the total number of hippocampal neurons observed in 10 dif- ferent populations likely has a genetic basis. Alterna- tively, it is also possible that the same selective agents (climate in our case) might have resulted in the evolution of the same phenotype with different underlying mechanisms (e.g. genetically based inheri- tance vs. experience-dependent plasticity). In our current study, we ask whether population genetic differentiation and potential isolation among populations is associated with large differences in hip- pocampal morphology. It has often been suggested that genetic isolation with resulting restriction of gene flow is important for the evolution of adaptive traits (e.g. Puechmaille et al. 2011), but it is also important to understand whether strong selection for particular traits may produce adaptive changes in the presence of gene flow in otherwise genetically undifferentiated popula- tions (Lenormand 2002). Therefore, we asked whether chickadee populations with large differences in hippo- campal attributes are genetically differentiated at neu- tral genetic loci and whether hippocampal differences are only observed among populations that are geo- graphically and genetically isolated. If patterns of genetic differentiation at neutral loci are associated with patterns of phenotypic differentiation, this may suggest that either isolation is necessary for adaptive divergence or phenotypic differences might be a product of genetic drift rather than of natural selection. Finding different or contrasting patterns of genetic and phenotypic differ- entiation, on the other hand, would suggest that the observed phenotypic differences could be a product of natural selection acting in the face of gene flow. These conclusions would be further strengthened if environ- Published 2012. This article is a US Government work and is in the public domain in the USA. CHICKADEE POPULATION STRUCTURE 4487 mental differences (winter climate), rather than genetic divergence at neutral markers, predicted the observed phenotypic differences. We used mitochondrial DNA sequences (mtDNA; Avise 2000) and amplified fragment length polymor- phism (AFLP; Vos et al. 1995; Pritchard et al. 2000; Falush et al. 2007) to test for genetic differentiation among the chickadee groups used in our previous stud- ies of spatial memory and the hippocampus. Mitochon- drial data have been widely used in phylogeographic studies and provide a useful comparison as a mater- nally inherited locus with a smaller effective population size compared to nuclear loci (Avise 2000). AFLPs have been used to estimate population structure because they can assay hundreds of anonymous markers sampled from throughout the genome, the vast majority of which are likely to be neutral (Bensch & A˚kesson 2005). The overall goal of our study was to investigate whether significant population divergence in hippocam- pal morphology is associated with genetic differentia- tion at neutral markers or whether divergence in hippocampal morphology is independent of genetic structure or geographic distance, but rather associated with differences in winter climate, which is expected to provide strong selection pressure on memory and the hippocampus. Materials and methods Samples We used the following numbers of birds from the 10 sam- pling locations used in our previous studies of the hippo- campus: Maine—12; Alaska, Fairbanks—34; AKA—34; British Columbia—37; Montana—40; Washington—33; Minnesota—12; Iowa—12; KS—24; and Colorado—19. Samples contained all of the birds that were those used for previous brain analyses including hippocampal vol- ume and the total number of hippocampal neurons (Table 1, Fig. 1; Roth & Pravosudov 2009; Roth et al. 2011) plus additional birds in most populations. It is important to note that there were no significant differ- ences between males and females in any of the measured brain parameters (Roth & Pravosudov 2009; Roth et al. 2011). DNA was extracted from blood using the mouse tail protocol on an AutoGenprep 965 extraction system (Autogen). DNA concentration and purity were assessed using a NanoDrop ND-1000 spectrophotometer. Fig. 1 A map of sampling locations (AKF, Alaska Fairbanks; AKA, Alaska Anchorage; BC, British Columbia, Prince George; MT, Montana, Missoula; WA, Washington, Seattle; CO, Colorado, Fort Collins; Kansas, Manhattan; IA, Iowa, Mt. Vernon; MN, Minnesota, Minneapolis; ME, Maine, Presque Isle). Unique shapes identify genetically distinct populations defined by STRUCTURE analysis. Letters within each shape correspond to the coarse comparative grouping based on hippocampal data: L, generally larger hippocampus with more neurons; S, generally smaller hippocampus with fewer neurons; I, somewhat intermediate values between L and S (based on Roth & Pravosudov 2009; Roth et al. 2011). Published 2012. This article is a US Government work and is in the public domain in the USA. 4488 V. V. PRAVOSUDOV ET AL. Mitochondrial DNA A portion of the mitochondrial control region was amplified by PCR using HCRCBox and LmochCR1 primers (Grava et al. 2012). Amplified DNA was sequenced by cycle sequencing using an ABI 3130xl sequencer. Chromatograms were visually inspected using MEGA V. 4.0.2 (Tamura et al. 2007). Mitochondrial haplotype diversity was estimated in DNASP V.5 (Librado & Rozas 2009). To investigate the distribution of genetic variation among populations, pairwise ΦST values were calculated using ARLEQUIN V. 3.5.1.2 (Excoffier & Lischer 2010) and significance was determined with 10 000 per- mutations. ΦST values are similar to F-statistics, except that the latter account for variation in allele frequencies, while the former encompass variation in frequencies as well as sequence divergence between haplotypes (i.e. all alleles are not equidistant). Insertion/deletions were treated as a fifth state. To correct for multiple tests, a Benjamini–Hochberg correction (Benjamini & Yekutieli 2001) was applied. To visualize the relationships among mitochondrial sequences, a haplotype network was con- structed using TCS V.1.21 using the default setting of 95% confidence limits (Clement et al. 2000). AFLP generation and scoring AFLPs were generated following the protocol of Vos et al. (1995) as modified for vertebrates by Kingston & Rosel (2004). Sixteen selective PCR primer pairs were used, consisting of all combinations of two fluorescently labelled EcoR1 + ANN primers and eight Taq1 + ANN primers. Fragment profiles were generated by capillary electrophoresis on an ABI PRISM 3730xl Genetic Ana- lyzer (Applied Biosystems). Fragment scoring was made using GENEMAPPER 4.0 software (Applied Biosys- tems). All samples were scored concurrently and blindly for each selective primer pair. AFLP fragments were grouped in 1-bp bins ranging in size from 90 to 500 bp. The scoring protocol developed by Kingston & Rosel (2004) was used to minimize potential noise asso- ciated with underamplification of large fragments or uneven amplification among samples. A threshold of 100 fluorescence units was set as the minimum ampli- tude for fluorescence peaks to qualify as potential mar- ker loci, while baseline fluorescence was generally below 25 units. The presence of false-negative peaks (<100 fluorescence units) in a bin with scorable peaks from other individuals resulted in the rejection of the marker. For each primer pair, the fragment length of the largest monomorphic marker was taken as the upper size limit for scorable loci to prevent scoring problems resulting from PCR drop off with fragment length.T a b le 1 M o rp h o lo g ic al an d b ra in m ea su re m en ts fr o m ch ic k ad ee s fr o m al l sa m p le d lo ca ti o n s (m ea n an d S E , n = 12 fo r ea ch lo ca ti o n ) L o ca ti o n B o d y M as s (g ) W in g le n g th (m m ) B ra in m as s (g ) A b so lu te h ip p o ca m p al v o lu m e (m m 3 ) A b so lu te n u m b er o f h ip p o ca m p al n eu ro n s (9 10 6 ) T el en ce p h al o n v o lu m e (m m 3 ) R el at iv e h ip p o ca m p al v o lu m e (m m 3 )* R el at iv e n u m b er o f h ip p o ca m p al n eu ro n s (9 10 6 )* D at es co ll ec te d A la sk a F ai rb an k s 11 .3 7 (0 .1 7) 65 .8 8 (0 .7 0) 0. 77 (0 .0 1) 28 .1 5 (0 .8 7) 1. 95 2 (0 .1 61 ) 49 4. 26 (1 5. 55 ) 28 .3 8 (0 .8 5) 2. 03 0 (0 .1 28 ) 18 –2 0 S ep te m b er 20 07 B ri ti sh C o lu m b ia 10 .8 7 (0 .1 7) 67 .0 5 (0 .6 4) 0. 80 (0 .0 2) 26 .6 0 (0 .8 2) 1. 87 7 (0 .1 26 ) 50 2. 61 (2 1. 94 ) 27 .0 4 (0 .9 8) 2. 01 3 (0 .1 48 ) 29 S ep te m b er –1 O ct o b er 20 07 M o n ta n a 11 .4 2 (0 .3 0) 65 .8 3 (0 .5 8) 0. 73 (0 .0 2) 25 .0 7 (0 .7 7) 1. 73 5 (0 .1 04 ) 44 6. 90 (2 1. 90 ) 25 .1 9 (0 .8 5) 1. 74 5 (0 .1 28 ) 10 O ct o b er 20 07 C o lo ra d o 11 .6 5 (0 .1 9) 67 .1 9 (0 .4 2) 0. 76 (0 .0 1) 25 .0 3 (0 .8 7) 1. 72 7 (0 .1 15 ) 48 5. 82 (1 1. 84 ) 25 .0 4 (0 .7 9) 1. 73 6 (0 .1 20 ) 15 –1 7 O ct o b er 20 07 K an sa s 12 .8 4 (0 .2 4) 69 .1 5 (0 .7 3) 0. 74 (0 .0 2) 23 .1 5 (0 .6 3) 1. 57 8 (0 .0 98 ) 45 9. 10 (1 4. 91 ) 22 .5 0 (1 .0 1) 1. 39 3 (0 .1 53 ) 21 –2 3 O ct o b er 20 07 A la sk a A n ch o ra g e 11 .1 0 (0 .1 6) 66 .0 4 (0 .7 6) 0. 80 (0 .0 2) 27 .2 0 (0 .6 4) 2. 58 2 (0 .0 65 ) 51 1. 41 (1 8. 36 ) 27 .3 8 (0 .7 3) 2. 60 3 (0 .0 94 ) 18 –2 0 S ep te m b er 20 08 W as h in g to n 11 .1 9 (0 .2 4) 61 .6 9 (0 .6 3) 0. 70 (0 .0 1) 22 .6 5 (0 .6 6) 1. 71 9 (0 .1 05 ) 43 0. 25 (1 4. 03 ) 22 .3 3 (0 .7 4) 1. 68 0 (0 .0 95 ) 22 O ct o b er 20 08 M ai n e 11 .6 0 (0 .2 3) 67 .5 0 (0 .4 8) 0. 78 (0 .0 2) 26 .0 8 (0 .9 9) 2. 53 7 (0 .1 06 ) 49 7. 13 (2 2. 85 ) 26 .1 7 (0 .7 2) 2. 54 8 (0 .0 92 ) 28 S ep te m b er 20 08 M in n es o ta 11 .6 4 (0 .2 5) 65 .5 4 (0 .6 1) 0. 78 (0 .0 1) 25 .9 1 (0 .7 2) 2. 56 0 (0 .0 92 ) 49 7. 20 (1 3. 34 ) 26 .0 0 (0 .7 2) 2. 57 0 (0 .0 92 ) 9 O ct o b er 20 08 Io w a 12 .1 1 (0 .2 6) 65 .0 8 (0 .7 5) 0. 74 (0 .0 2) 24 .0 6 (0 .4 5) 1. 60 9 (0 .0 76 ) 48 2. 53 (1 9. 88 ) 24 .0 6 (0 .7 2) 1. 60 9 (0 .0 91 ) 13 –1 6 O ct o b er 20 08 *L ea st -s q u ar e m ea n s (b as ed o n R o th & P ra v o su d o v 20 09 ; R o th et al . 20 11 ). Published 2012. This article is a US Government work and is in the public domain in the USA. CHICKADEE POPULATION STRUCTURE 4489 AFLP data were analysed using the Bayesian popu- lation clustering program STRUCTURE (v. 2.3.2) (Prit- chard et al. 2000; Falush et al. 2007). Prior to running STRUCTURE, we visually examined the data for statistical or artefactual outliers using nonmetric multidimen- sional scaling (NMDS) and detected no problematic individuals. In STRUCTURE, we ran an admixture model with correlated alleles 10 times for each value of k (from 2 to 15) with burn-in of 100 000 steps followed by sampling of 1 000 000 steps. To determine the most informative value of k, we used the ad hoc sta- tistic Dk following Evanno et al. (2005). In addition, we also compared variance at each k to test whether an increase in variance occurs after the k with the largest Dk. After determining the most effective level of k for our data, we re-ran STRUCTURE using the USE- POPINFO option to incorporate sampling locations as prior information, which has been suggested as a method for inferring direct gene flow in recent gener- ations subsequent to the identification of population structure (Pritchard et al. 2000; Falush et al. 2007). We also examined STRUCTURE results for the k-values below the largest Dk value. Finally, the program AFLP-SURV was used with default settings to generate pairwise Fst values between all populations (Vekemans et al. 2002), which were used in multiple Mantel tests as described below. Comparisons between genetic distances and phenotypic traits Mantel and partial Mantel tests have often been used to investigate the relationships among populations associ- ated with genetic distance, geographic distance and phenotypic or ecological traits (Legendre & Legendre 1998). More recently, a multiple regression technique has been proposed by Lichstein (2007), which allows for the simultaneous consideration of multiple predictor variables while retaining the permutations appropriate for the nonindependence of distance matrices observed in a spatial framework. We used this approach as implemented in the R package ecodist to model differ- ences in hippocampal morphology among populations as predicted by genetic divergence (Fst values derived from AFLPs), geographic distance and differences in winter climate (winter temperature). Temperature was averaged (prior to the creation of a distance matrix based) over winter months (November–March) tabu- lated for the period 1971–2000 (Roth & Pravosudov 2009; Roth et al. 2011). For hippocampal traits, we used both population mean values for relative hippocampal volume and the total number of hippocampal neurons (Table 1). The data were z-transformed to generate stan- dardized values before the creation of pairwise differ- ences used in multiple Mantel analysis. For both hippocampal parameters and winter temperature aver- aged over four winter months, distance matrices included in multiple Mantel tests were based on Euclid- ean distances, calculated with the dist function in R. Results Mitochondrial DNA A total of 46 variable sites, including two insertion/dele- tion events, defined 51 haplotypes in a 675-bp sequence of the mitochondrial control region in 146 birds (Alaska, Fairbanks—12, AKA—12, British Columbia—12, Wash- ington—11, Montana—22, Colorado—19, Iowa—12, KS —23, Minnesota—12, Maine—11). Twenty-one of the haplotypes were found in more than one individual, and ten of the shared haplotypes were restricted to a single sampling site. Haplotype diversity was high, ranging from 0.46 to 0.93, and was generally higher in the eastern populations (Kansan, Minnesota and Maine). The number of unique haplotypes, those found in a sin- gle bird, was lower in the west than in the east. Pairwise ΦST values ranged from 0.027 to 0.656. The ΦST values were significant for 37 of the 45 pairwise tests indicating genetic differences among populations (Table 2). AKA and Alaska, Fairbanks were not significantly different from each other, but were significantly different from all the other sampled populations, and had lower haplo- type diversity. Similarly, Washington was significantly different from all other populations and had low haplo- type diversity. Colorado had high haplotype diversity and was significantly different from all other sampled populations, although it did share haplotypes with birds from adjacent sampling sites in Montana and KS. The generally high level of differentiation in mitochondrial sequences (Table 2) can also be seen in the haplotype network (Fig. 2). Few haplotypes are found in more than one or even two populations, and the distribution of haplotypes shows some geographic structure (note the clustering, for example, of Alaskan sequences). The dis- tribution of haplotypes is indicative of both gene flow and a lack of differentiation in some areas. Note, for example, some of the more interior haplotypes in the network that are shared among multiple populations, such as the single haplotype that is found in Iowa, KS, Minnesota and Maine (Fig. 2B). AFLP The AFLP analyses were based on samples from 257 birds and generated 583 loci (0.16% of missing data). There were 167.03 (mean with SD = 6.33) bands present Published 2012. This article is a US Government work and is in the public domain in the USA. 4490 V. V. PRAVOSUDOV ET AL. per individual, and we have found in total 547 poly- morphic bands and 36 fixed bands. STRUCTURE analysis of AFLP loci showed that ln(likeli- hood) continuously increased as k increased, consistent with some genetic structure among our study popula- tions. Using Dk estimation, we determined that four clusters provided the best fit to the data: (i) an Alaska population that includes birds from both Anchorage and Fairbanks; (ii) a British Columbia and Montana population; (iii) a Washington population; and (iv) a population cluster including Maine, Minnesota, Iowa, KS and Colorado (Fig. 2A). All individuals in Alaska, Washington, KS and Iowa were assigned to their respective populations with very few individuals hav- ing small probabilities of being assigned to any other population. A few more birds assigned to the British Columbia–Montana cluster had some probability of being assigned to either Alaska, Washington or the Maine–Minnesota–Iowa– KS –Colorado cluster. Finally, a few birds from Maine, Minnesota and Colorado had some probabilities of being assigned to the British Columbia–Montana (mainly), Washington or Alaska populations. Nonetheless, all four populations showed robust genetic structure suggesting significant genetic differentiation (Fig. 2A). Variance for the mean value of ln likelihood among all runs was low for k ranging from 2 to 4 (SD = 0.3, 0.7, 0.5, respectively), but rising dramatically at k = 5 (SD = 61.2), which also supports our conclusion that the data are well partitioned into four genetically differentiated clusters. Although the best fit was predicted with k = 4, the results of STRUCTURE analyses at lower values of k are also supportive of the separation of the two Alaska populations as well as Washington, British Columbia and Montana into different clusters. At k = 2, STRUCTURE separates both Alaska groups and all other locations into two separate clusters, and at k = 3, Alaska Anchor- age and Alaska Fairbanks form one cluster, Washing- ton, British Columbia and Montana form the second cluster, and all other locations (Colorado, KS, Iowa, Minnesota, Maine) form the final third cluster. After establishing that all of our birds most likely form four genetically distinct populations, we used the USEPOPINFO option (Pritchard et al. 2000; Falush et al. 2007) to investigate the probability that some of the individuals in these four clusters may have an immi- grant mixed ancestry in the last three generations (set in the program by the GENSBACK option), which would be suggestive of direct gene flow between these popula- tions. We used GENSBACK = 3 generations and MIGRPRI- OR = 0.05 (0.95 probability that the individual has pure ancestry from the assigned population). This model showed that no individuals in any of the four desig- nated populations have a significant probability of being of mixed immigrant ancestry during the last three generations, suggesting low or potentially no direct gene flow between these four populations, or at least between the exact locations representing these popula- tions (Fig. 2A). Because the inference of no genetic structure among eastern populations is important for the conclusions that we draw, we conducted an additional, focused analysis of five eastern groups predicted to form the same genetically undifferentiated cluster (Iowa, Minne- sota, Maine, KS and Colorado) to test whether a sepa- rate analysis of just these groups might detect a subtle genetic structure that may have been missed in the pre- vious larger analysis including all 10 groups. STRUCTURE predicted no genetic structure within these groups at all tested values of k (2 through 7), with all birds having high probabilities to form only a single genetically distinct group. These results support the main model, which placed these five groups into a single genetically undifferentiated cluster. It is important to note that the geographic distance between some of the eastern populations is much larger than that between, for exam- Table 2 mtDNA pairwise ΦST values (below diagonal) and P values (above) AKF AKA BC WA MT CO IA KS MN ME AKF * 0.276 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 AKA 0.027 * <0.0001 0.0001 <0.0001 <0.0001 <0.0001 <0.0001 0.0001 <0.0001 BC 0.615 0.518 * <0.0001 0.017 <0.0001 0.045 <0.0001 0.104 0.082 WA 0.450 0.289 0.425 * 0.001 <0.0001 <0.0001 0.003 0.0004 <0.0001 MT 0.438 0.344 0.118 0.244 * 0.028 0.104 0.002 0.105 0.002 CO 0.489 0.375 0.288 0.282 0.066 * 0.001 0.035 0.007 <0.0001 IA 0.610 0.495 0.108 0.375 0.048 0.143 * 0.0004 0.112 0.027 KS 0.487 0.366 0.285 0.193 0.114 0.059 0.195 * 0.048 <0.0001 MN 0.530 0.417 0.056 0.287 0.043 0.092 0.055 0.070 * 0.021 ME 0.656 0.562 0.074 0.470 0.157 0.303 0.109 0.299 0.090 * AKA, Alaska Anchorage; AKF, Alaska Fairbanks. Values significant after correction for multiple tests (Pcrit = 0.0452) are indicated in boldface. Published 2012. This article is a US Government work and is in the public domain in the USA. CHICKADEE POPULATION STRUCTURE 4491 ple, Iowa, Minnesota and Montana, but the Montana population has been detected as genetically distinct from both Iowa and Minnesota. Finally, a multiple Mantel test significantly predicted differences in hippocampal morphology among popula- tions based on the inclusion of winter climate (tempera- ture), geographic distance and pairwise Fst values derived from AFLP variation; results from the multiple Mantel test were as follows: R2 = 0.31, F = 6.19, and P = 0.008. Among the included variables, only winter climate was a significant, individual predictor of popu- lation differences in hippocampal morphology (Fig. 3); (B) (A) Fig. 2 (A) Population assignment of chickadees from 10 geographic locations (Fig. 1) based on STRUCTURE analysis of AFLP at k = 2, 3 and 4 (top three panels) and likely ancestry of individuals in four identified populations at k = 4, estimated using USEPOPINFO option in STRUCTURE (bottom panel). (B) Haplotype distribution about 10 geographic locations. Numbers after locations refer to how many individuals share that haplotype (e.g. MT3, three birds from MT have the same haplotype). Small white circles represent inferred haplotypes. The size of the ovals is proportional to the number of individuals possessing a given haplotype. Colours of all clusters and haplotypes match the colours of the STRUCTURE-assigned populations at k = 4. Published 2012. This article is a US Government work and is in the public domain in the USA. 4492 V. V. PRAVOSUDOV ET AL. in contrast, both the effects of geographic distance and Fst values were nonsignificant (Fig. 3). Although for simplicity we have focused on a single, multiple Mantel test involving three predictor variables (Lichstein 2007), we also explored pairwise analyses between each pre- dictor variable and population differences in hippocam- pal morphology. Also, we investigated partial Mantel tests between population differences in hippocampal morphology and both differences in winter climate and AFLP Fst values, while accounting for geographic dis- tances. In all of these cases, winter climate emerges clearly as being most strongly and significantly associ- ated with population divergence in hippocampal mor- phology (P < 0.05), while the effects of geographic distance and neutral genetic divergence remain nonsig- nificant. Comparing mtDNA and AFLP results AFLP and mtDNA analyses reached generally similar conclusions in that they both separated both Alaska populations and Washington into genetically distinct populations, and both groups of analyses found evi- dence consistent with gene flow among the eastern pop- ulations. This can be seen in the ΦST values (Table 1) and also in the mitochondrial genealogy: AKA and Alaska Fairbanks haplotypes in particular were mostly unique (one haplotype was shared with Montana, and one was shared with Montana, Washington and British Columbia birds) and occupy a peripheral position in the network, suggesting a more derived position rela- tive to the lower, continental populations (which are also likely larger populations, given the higher haplo- type diversity and greater number of birds with unique haplotypes). Iowa and Minnesota birds were also pre- dicted to be genetically similar by both analyses. Unlike the results from AFLPs, mtDNA analyses made more significant distinctions among some of the other groups. For example, Colorado birds were significantly different from birds from all other locations (Table 1). KS birds were similar to the birds from Minnesota, but were sig- nificantly different from birds from all other locations. Chickadees from British Columbia were similar to chickadees from Iowa, Minnesota and Maine, but differ- ent from all other locations. Finally, Montana chicka- dees were similar to chickadees from Iowa and Minnesota, but distinct from chickadees from all other locations. One specific difference between the AFLP and the mtDNA analyses concerned birds from British Columbia and Montana: STRUCTURE was unable to detect differences between birds from the two sampling sites, while the ΦST analyses of mitochondrial variation showed significant differences between these two groups (unique haplotypes in both areas are also evi- dent in the haplotype network, Fig. 2B). Overall, how- ever, a pairwise Mantel test using ΦST values from mtDNA analysis and Fst from AFLP analysis showed a significant association (correlation coefficient = 0.76, P = 0.006). Given that mtDNA is a maternally inherited marker, sex-bias in dispersal patterns may potentially affect the results of the analyses. Unfortunately, no detailed infor- mation exists about sex-specific dispersal in the black- capped chickadee, but it is likely that, as in most bird species, it is female-biased. However, significant simi- larity of the mtDNA and AFLP analyses suggests that our results are fairly robust. Discussion The population genetic results reported here show genetically structured chickadee populations through- 0 5 10 15 20 25 0. 0 0. 5 1. 0 1. 5 2. 0 2. 5 3. 0 3. 5 Temperature (C) H ip po ca m pu s (v ol um e an d # ne ur on s) P = 0.0018 (A) 0.00 0.02 0.04 0.06 0.08 0.10 0. 0 0. 5 1. 0 1. 5 2. 0 2. 5 3. 0 3. 5 Fst (AFLPs) H ip po ca m pu s (v ol um e an d # ne ur on s) P = 0.35 (B) 1000 2000 3000 4000 5000 0. 0 0. 5 1. 0 1. 5 2. 0 2. 5 3. 0 3. 5 Geographic distance (kms) H ip po ca m pu s (v ol um e an d # ne ur on s) P = 0.33 (C) Fig. 3 Associations among pairwise population differences in hippocampal morphology (volume and the total number of neurons, z- transformed data) and (A) pairwise differences in winter climate (ambient mean daily temperature averaged over winter months— November–March), (B) pairwise genetic divergence (AFLP Fst) and (C) pairwise geographic distance. The P values associated with each graph are from a multiple Mantel test involving all three variables (see text for details). Published 2012. This article is a US Government work and is in the public domain in the USA. CHICKADEE POPULATION STRUCTURE 4493 out North America. However, while some populations with known differences in hippocampal morphology have been found to be genetically distinct, others appear to be genetically undifferentiated, suggesting relatively unrestricted gene flow. Overall, population divergence in phenotypic traits (hippocampal morphol- ogy) is not associated with genetic differentiation at neutral markers, which is consistent with the possibility that, at least in some populations, these differences have evolved in response to natural selection and potentially in the face of gene flow. Pairwise analyses of population genetic divergence, geographic distance and winter climate further support this conclusion by showing that only winter climate is significantly associ- ated with population phenotypic divergence in hippo- campal morphology. Analyses of AFLPs identified four regional genetic clusters (Fig. 2A) as having the highest likelihood for explaining the variation revealed in our data. In partic- ular, birds from the two Alaska populations and Wash- ington were quite distinct, both from each other and from other populations. Specimens from British Colum- bia and Montana were identified as deriving from one population that was also fairly distinct from all other areas, even though the Montana population contains a large diversity of mitochondrial haplotypes. Impor- tantly, the variation in mitochondrial DNA was broadly consistent with a picture of population structure and some level of population isolation: many populations contained unique haplotypes (Fig. 2B) and differences inferred by ΦST analyses found the majority of popula- tions to be distinct, even within clusters identified by the AFLP analyses. Significant genetic differentiation suggests that the gene flow between Alaska, Washington, British Colum- bia–Montana and the rest of the locations may be lim- ited relative to the gene flow within each of these groups. At the same time, our nuclear data strongly suggest that gene flow is comparatively unrestricted between Iowa, Maine, Minnesota, KS and likely Colorado. We stress that these inferences should be considered in a comparative context: that is, Maine and Minnesota have, for example, shared a more recent demographic history with each other than either of those areas has with Alaska or Washington. AFLP variation includes many loci randomly distributed throughout the genome, the vast majority of which are expected to be neutral and so differences in genetic structure likely reflect changes due to genetic drift. It is important to note that our sample sizes for Maine, Min- nesota and Iowa were slightly lower (12) than those for other populations and that these low sample sizes could theoretically be responsible for our failure to find signif- icant genetic differentiation between these populations. We, however, do not think this is the case, as some of the populations in the same genetically undifferentiated cluster (Colorado and KS) have larger sample sizes, and all of these populations were significantly different from Montana, British Columbia, Washington and Alaska populations. We cannot exclude the possibility that some of the loci sampled may not be neutral and, in fact, we would hypothesize that there should be significant dif- ferences in some genetic regions involved in hippo- campal functions. However, those unidentified regions associated with brain function (however important bio- logically) must comprise a relatively small proportion of the total genome and would therefore be highly unlikely to be included in our AFLP data set. More- over, even if functional genetic regions associated with the hippocampus were included, differences in these genes would produce a population genetic structure that parallels the differences in memory and hippo- campal morphology among our sampling locations. Our results, however, showed that differences in hip- pocampal morphology were significantly associated with differences in winter climate, but not with genetic divergence at neutral loci or geographic distance (Fig. 3). Of course, we cannot put a date on divergence in hip- pocampal traits among different populations; thus, we do not know whether hippocampal changes arose in the face of contemporary gene flow (e.g. among popula- tions within the red cluster, Fig. 2A) or whether they arose at an earlier time (while populations were more isolated) and persist today despite the gene flow. How- ever, both interpretations are equally interesting and suggestive of natural selection affecting the variation in ecological traits that has been observed. Our data suggest that winter climate is the overarch- ing factor associated with selection on hippocampal structure supporting the evolution of enhanced spatial memory, as hypothesized by previous studies in this system, and it seems likely that these traits respond to climate-based selection independent of gene flow. Chickadees from the two Alaska populations have lar- ger hippocampi containing more neurons compared to the birds from almost all other locations, except Maine and Minnesota and marginally British Columbia (Roth & Pravosudov 2009; Roth et al. 2011), and they also appear to be genetically distinct from all other populations (Fig. 1). On the other hand, chickadees from Minnesota and Maine have larger hippocampi containing more neurons compared to the birds from Iowa and KS (Roth & Pravosudov 2009; Roth et al. 2011), yet all four locations form a single genetically distinct population and likely share extensive gene flow, which is additionally supported by the mtDNA Published 2012. This article is a US Government work and is in the public domain in the USA. 4494 V. V. PRAVOSUDOV ET AL. analysis. Finally, chickadees from Washington are genetically distinct from all other populations, yet their hippocampal morphology is similar to that in birds from KS and Iowa, which share somewhat similar envi- ronment. A Mantel test in a multiple regression frame- work solidified these conclusions by showing that population divergence in hippocampal morphology is significantly associated with differences in winter cli- mate, but not in genetic divergence or geographic dis- tance (Fig. 3). We note that we have investigated population genetic structure, but have not explicitly addressed the possibil- ity for genetic clines associated with, for example, isola- tion by distance. This approach was necessitated by our sampling scheme, which was in turn driven by previ- ous experimental work. Thus, the results presented here do not rule out the possibility of more subtle clinal vari- ation within, for example, the eastern populations. A recent simulation study (Chen et al. 2007) concluded that STRUCTURE is highly capable of detecting clinal varia- tion in contrast to previous suggestions to the contrary. In addition, our secondary analysis with STRUCTURE that focused only on the eastern populations (the red group in Fig. 2) while excluding the other populations was consistent with the hypothesis that these populations form a single, large gene pool. Moreover, mtDNA results also suggest relatively less genetic differentiation at least between some of the populations showing the largest hippocampal differences (Minnesota and Iowa). In fact, birds from Minnesota and Iowa share haplo- types suggesting gene flow between these populations, either contemporary or in the recent past. In addition, STRUCTURE did detect genetic differentiation between Montana and Minnesota and Iowa and the geographic distance between them is smaller than between Iowa, Minnesota and Maine, which have all been assigned to a single population. Our previous work suggested that the differences in hippocampal morphology between the Alaska and KS populations, which are genetically differentiated, are not experience-based but rather inherited (Roth et al. 2012). In addition, our previous comparison of labora- tory-reared and maintained and wild chickadees strongly suggested that the variance in at least the total number of hippocampal neurons has a heritable basis rather than being caused by environment-dependent experiences (Roth et al. 2012). If we assume that the relationship between the hippocampus morphology and climate is the result of selection for spatial memory in all populations, then combined, these data suggest that strong selection on memory and hippocampal morphol- ogy imposed by winter climate severity may produce adaptive changes in these traits regardless of genetic differentiation and potential isolation among popula- tions and hence likely regardless of the relative amount of gene flow (and this inference holds even if traits arose at an earlier time, as discussed above). This would suggest that differences in memory and hippo- campal morphology might evolve independently in dif- ferent populations as a function of climate-induced selection on memory. Alternatively, it may also be possible that the mecha- nisms generating and maintaining significant variation in memory and hippocampal morphology may be dif- ferent between genetically differentiated and undiffer- entiated populations. While differences in hippocampal morphology between genetically differentiated popula- tions may have a heritable basis as a result of selection, similar differences between relatively undifferentiated populations (such as those hippocampal differences separating Iowa and Minnesota) may be maintained by experience-dependent plasticity. These possibilities can- not be addressed with the data in hand and must await the collection of further experimental data to differenti- ate between potential mechanisms. In addition to expe- rience-dependent plasticity, the observed differences among some of the populations may also be, at least in part, due to parental and epigenetic effects. The popula- tion genetic results that we have presented here lay the foundation for future studies that should focus specifi- cally on genetically undifferentiated populations in close geographic proximity in order to investigate whether variation in memory and hippocampal mor- phology may be maintained by different mechanisms, and how these mechanisms might interact with differ- ing levels of gene flow. Beyond the identification of population structure in the context of divergence in hippocampal traits, the mitochondrial results point towards a complex biogeo- graphic history for the black-capped chickadee in North America that is beyond the scope of this study. For example, Alaska has low diversity and may have been the result of a founder event, which could explain the structure of the haplotype network with geographically proximate populations (e.g. Alaska and British Colum- bia) not directly linked. In general, the presence of gla- cial refugia during Pleistocene climatic cycles might likely be reflected in the divergence of populations from Alaska, northwestern areas and central continental pop- ulations (Hewitt 2000). However, more geographically intensive sampling is needed to make inferences about biogeographic history; instead, our purpose here has been to place previous behavioural and morphological work in a population genetic context to set the stage for future investigations into natural selection and diversifi- cation in memory-related traits. Published 2012. This article is a US Government work and is in the public domain in the USA. CHICKADEE POPULATION STRUCTURE 4495 Acknowledgements The project was partially funded by NSF IOB-0615021 to VVP and NSF IOB-0918268 to LDA and VVP. Work was approved by the University of Nevada IACUC #A05/06-35 and followed all federal and local guidelines for the use of animals in research. We thank Ken Otter and Angelique Grava for samples from UNBC. TMB would like to thank John Hindley and Linda Lait for assistance in the laboratory, and Heather Bird, Karley Campbell, Brendan Graham, John Hindley, Elysha Koran, Linda Lait, Christine Michell and Paulo Pulgarin for the field work obtaining samples. TMB was funded by NSERC DG. References Avise JC (2000) Phylogeography: The History and Formation of Species. Harvard University Press, Cambridge, MA, USA, 464p. Benjamini Y, Yekutieli D (2001) The control of the false discov- ery rate in multiple testing under dependency. Annals of Sta- tistics, 29, 1165–1188. Bensch S, A˚kesson M (2005) Ten years of AFLP in ecology and evo- lution: why so few animals?Molecular Ecology, 14, 2899–2914. Chancellor LV, Roth II TC, LaDage LD, Pravosudov VV (2011) The effect of environmental harshness on neurogenesis: a large scale comparison. Developmental Neurobiology, 71, 246–252. Chen C, Durand E, Forbes F, Francois O (2007) Bayesian clus- tering algorithms ascertaining spatial population structure: a new computer program and a comparison study. Molecular Ecology Notes, 7, 749–756. Clement M, Posada D, Crandall K (2000) TCS: a computer program to estimate gene genealogies. Molecular Ecology, 9, 1657–1660. Evanno G, Regnaut S, Goudet J (2005) Detecting the number of clusters of individuals using the software structure: a simula- tion study. Molecular Ecology, 14, 2611–2620. Excoffier L, Lischer HEL (2010) Arlequin suite ver 3.5: a new series of programs to perform population genetics analyses under Linux and Windows. Molecular Ecology Resources, 10, 564–567. Falush D, Stephens M, Pritchard JK (2007) Inference of popula- tion structure using multilocus genotype data: dominant markers and null alleles. Molecular Ecology Notes, 7, 574–578. Grava A, Grava T, Didier R et al. (2012) Interspecific domi- nance and hybridization between black-capped and moun- tain chickadees. Behavioral Ecology, 23, 566–572. Hampton RR, Shettleworth SJ (1996) Hippocampal lesions impair memory for location but not color in passerine birds. Behavioral Neuroscience, 110, 831–835. Hewitt G (2000) The genetic legacy of the Quaternary ice ages. Nature, 405, 907–913. Kingston SE, Rosel PE (2004) Genetic differentiation among recently diverged delphinid taxa determined using AFLP markers. Journal of Heredity, 95, 1–10. Krebs JR, Sherry DF, Healy SD, Perry VH, Vaccarino AL (1989) Hippocampal specialization of food-storing birds. Proceedings of the National Academy of Sciences of the United States of Amer- ica, 86, 1388–1392. Legendre P, Legendre L (1998) Numerical Ecology, 2nd edn. Elsevier, New York City, New York. Lenormand T (2002) Gene flow and the limits to natural selec- tion. Trends in Ecology & Evolution, 17, 183–189. Librado P, Rozas J (2009) DnaSP v5: a software for comprehen- sive analysis of DNA polymorphism data. Bioinformatics, 25, 1451–1452. Lichstein JW (2007) Multiple regression on distance matrices: a multivariate spatial analysis tool. Plant Ecology, 188, 117–131. Orr HA (2005) The genetic theory of adaptation: a brief history. Nature Review Genetics, 6, 119–127. Pravosudov VV, Clayton NS (2002) A test of the adaptive spe- cialization hypothesis: population differences in caching, memory and the hippocampus in black-capped chickadees (Poecile atricapilla). Behavioral Neuroscience, 116, 515–522. Pravosudov VV, Smulders TV (2010) Integrating ecology, psy- chology, and neurobiology within a food-hoarding para- digm. Philosophical Transactions of the Royal Society of London. Series B. Biological Science, 365, 859–867. Pritchard JK, Stephens M, Donnelly P (2000) Inference of popu- lation structure using multilocus genotype data. Genetics, 155, 945–959. Puechmaille SJ, Gouilh MA, Piyapan P et al. (2011) The evolu- tion of sensory divergence in the context of limited gene flow in the bumblebee bat. Nature Communications, 2, 1–9. Roth II TC, Pravosudov VV (2009) Hippocampal volume and neuron numbers increase along a gradient of environmental harshness—a large-scale comparison. Proceedings of the Royal Society of London. Series B, 276, 401–405. Roth II TC, LaDage LD, Pravosudov VV (2011) Variation in hippocampal morphology along an environmental gradient: controlling for the effect of day length. Proceedings of the Royal Society of London. Series B, 278, 2662–2667. Roth II TC, LaDage LD, Freas C, Pravosudov VV (2012) Var- iation in memory and the hippocampus across populations from different climates: a common garden approach. Pro- ceedings of the Royal Society of London. Series B, 279, 402– 410. Sherry DF, Vaccarino AL (1989) Hippocampus and memory for food caches in black-capped chickadees. Behavioral Neuro- science, 103, 308–318. Sherry DF, Vaccarino AL, Buckenham K, Hertz RS (1989) The hippocampal complex of food-storing birds. Brain Behavior and Evolution, 34, 308–317. Shettleworth SJ (1995) Memory in food-storing birds: from the field to the skinner box. In: Behavioral Brain Research in Naturalistic and Semi-naturalistic Settings (eds Alleva E, Fa- solo A, Lipp H-P, Nadel L), pp. 158–179. Proc. NATO Advanced Study Institute Series Maratea, Italy, Kluwer Aca- demic Publishers, The Hague, The Netherlands. Tamura K, Dudley J, Nei M, Kumar S (2007) MEGA4: Molec- ular Evolutionary Genetics Analysis (MEGA) software ver- sion 4.0. Molecular Biology and Evolution, 24, 1596–1599. Vander Wall SB (1990) Food Hoarding in Animals. University of Chicago Press, Chicago, Illinois. Vekemans X, Beauwens T, Lemaire M, Roldan-Ruiz I (2002) Data from amplified fragment length polymorphism (AFLP) markers show indication of size homoplasy and of a rela- tionship between degree of homoplasy and fragment size. Molecular Ecology, 11, 139–151. Published 2012. This article is a US Government work and is in the public domain in the USA. 4496 V. V. PRAVOSUDOV ET AL. Vos P, Hogers R, Bleeker M et al. (1995) AFLP: a new tech- nique for DNA fingerprinting. Nucleic Acids Research, 23, 4407–4414. V.V.P., T.C.R., L.LD.: interested in cognitive ecology, neuroe- cology, evolution of cognition, genetics. M.L.F.: interested in evolutionary ecology and population genetics. T.M.B., M.J.B., B.S.D.: interested in population genetics. Data accessibility mtRNA: GenBank accessions JX164148–JX164198. AFLP: Dryad accession doi:10.5061/dryad.5p6b1. GenBank sequences: supplemental material available with the online version of this article. Supporting information Additional Supporting Information may be found in the online ver- sion of this article. Appendix S1 mtDNA GenBank sequences. Please note: Wiley-Blackwell are not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article. Published 2012. This article is a US Government work and is in the public domain in the USA. CHICKADEE POPULATION STRUCTURE 4497